JP7462424B2 - Medical information processing device, learning data generation program, and learning data generation method - Google Patents

Description

This embodiment relates to a medical information processing device, a learning data generation program, and a learning data generation method.

Traditionally, radiology doctors who interpret images select training data in addition to their regular interpretation duties. At this time, the doctors must add appropriate information to the selected training data. In addition, the doctors write down the results of the interpretation in an interpretation report, which is the output of the interpretation. However, the results of the interpretation are not always correct. In such cases, the doctors must check medical documents such as pathology test results and medical records.

Checking these data annotations and diagnosis results one by one while carrying out busy work tasks places a heavy burden on radiologists. In other words, creating correct answer data for the training data requires a great deal of effort for radiologists.

JP 2017-204041 A

The problem that this invention aims to solve is to generate training data easily.

The medical information processing device according to this embodiment includes an identification unit, an acquisition unit, an extraction unit, and an integration unit. The identification unit identifies the examination purpose of the first examination based on a first examination result related to a medical image acquired by a first examination on a subject. The acquisition unit acquires a second examination result of a second examination performed on the subject for the examination purpose from a medical data server based on order information related to the first examination result. The extraction unit extracts first examination data usable for learning a machine learning model from the first examination result, extracts second examination data usable for the learning from the second examination result, and extracts a learning image usable for the learning from the medical image. The integration unit integrates the first examination data, the second examination data, and the learning image as learning data usable for the learning.

FIG. 1 is a diagram showing an example of the configuration of a medical information processing apparatus according to an embodiment, together with a hospital information system. FIG. 2 is a diagram showing an example of a relationship between an acquisition target by an acquisition function and an extraction target by an extraction function in the embodiment. FIG. 3 is a diagram showing an example of learning data generated by integrating various data by an integration function according to the embodiment. FIG. 4 is a diagram showing an example of a processing procedure of a learning data generation process according to the embodiment.

Hereinafter, with reference to the drawings, a detailed description will be given of embodiments of a medical information processing device, a learning data generation program, and a learning data generation method. In the following embodiments, parts with the same reference numerals perform similar operations, and duplicated descriptions will be omitted as appropriate. Note that the medical information processing device according to the present application is not limited to the embodiments shown below.

(Embodiment)
The overall configuration of a medical information processing device according to an embodiment will be described. FIG. 1 is a diagram showing an example of the configuration of a medical information processing device 1 according to an embodiment together with a hospital information system (HIS: Hospital Information System) 3. The medical information processing device 1 is connected to the hospital information system 3 via a communication network 31. First, the hospital information system 3 related to the medical information processing device 1 will be described, and then the medical information processing device 1 will be described. The hospital information system 3 has an electronic medical record system 33, a radiology department information system 35, a report creation device 36, a medical image diagnostic device 37, a medical image management system 39, a clinical examination department information system 41, and other department systems 43. The electronic medical record system 33, the radiology department information system 35, the report creation device 36, the medical image diagnostic device 37, the medical image management system 39, the clinical examination department information system 41, and other department systems 43 are connected to each other via a wired or wireless communication network 31.

The electronic medical record system 33 is an information system that manages electronic medical records for recording medical treatment details. The electronic medical record system 33 includes an electronic medical record server device and an electronic medical record terminal device. The electronic medical record server device is a computer device that executes processes related to the management of electronic medical records. The electronic medical record server device is composed of one to multiple computer devices depending on the scale of the system. The electronic medical record terminal device is used by doctors, nurses, etc. who input and refer to the electronic medical records, and one to multiple devices are provided depending on the scale of the system. The electronic medical record server device and the electronic medical record terminal device are connected to the communication network 31.

The electronic medical record terminal device inputs an examination order for a subject in response to a user's instruction. If the examination order is for imaging of the subject for image diagnosis, the examination order includes information on the examination content, such as the examination site (imaged subject, disease name, examination purpose, modality, etc. The examination order may also include information such as the patient ID, patient name, examination ID, and examination date of the subject undergoing the examination. In response to the input of the examination order, the electronic medical record server device transmits information on the examination order (hereinafter referred to as order information) to the radiology department information system 35 together with the patient information of the subject in the order information. If the examination order is for the performance of a clinical test on the subject, the electronic medical record server device transmits the order information to the clinical laboratory department information system 41 in response to the input of the examination order.

The test results for all tests listed in the order information are linked to the electronic medical record for the subject. At this time, the clinician who requested the test makes a diagnosis based on the test results and records the diagnosis in the electronic medical record for the subject. The diagnosis includes information such as the diagnosis name, stage, and severity of the disease. The stage and severity of the disease are defined by a predetermined definition document. The electronic medical record terminal device inputs the diagnosis results based on various tests on the subject, for example, at the instruction of the clinician. The diagnosis results are stored in the electronic medical record server device. In addition, the electronic medical record terminal device inputs the results of treatment (hereinafter referred to as treatment results) performed based on the diagnosis results, at the instruction of the clinician. The treatment results are stored in the electronic medical record server device. The treatment results include, for example, whether or not surgery was performed, the area to be removed from the surgical site (for example, total removal, partial removal), whether or not surgery is performed after laparotomy, whether or not radiation therapy is performed, and a series of treatment events in chronological order.

The radiology information system (RIS) 35 is an information system that manages information in the radiology department within the hospital, and is equipped with a radiology department information server device and a radiology department information terminal device. The radiology department information server device is a computer device that executes processes related to the management of information in the radiology department. The radiology department information server device is composed of one to multiple computer devices depending on the scale of the system. The radiology department information terminal device is used by technicians in the radiology department, and one to multiple devices are provided depending on the scale of the system. The radiology department information server device and the radiology department information terminal device are connected to the communication network 31.

As an example of the operation of the radiology department information server device, the radiology department information server device receives order information from the electronic medical record system 33 and acquires information on the medical imaging diagnostic device 37 from the order information. The radiology department information server device transmits the order information to the medical imaging diagnostic device 37 identified by the order information. When the order information is transmitted to the medical imaging diagnostic device 37, the radiology department information server device may add information such as the patient ID and examination date to a DICOM (Digital Imaging and Communications in Medicine) tag, etc.

The report creation device 36 is a device with which the image interpretation doctor creates an image interpretation report for a medical image. The report creation device 36 may be realized by a PC (Personal Computer) terminal in the medical image management system 39. The report creation device 36 receives the order information and patient information transmitted from the electronic medical record system 33 via the communication network 31. In response to receiving the order information and patient information, the report creation device 36 creates a template for an image interpretation report in which specific information such as patient information, examination information, and requester information that is the subject of the image interpretation report is written.

The report creation device 36 acquires medical images required for creating an image interpretation report from the medical image management system 39. When an instruction to create an image interpretation report is given, the report creation device 36 displays a report creation screen for a template image interpretation report in which specific information has been written. The report creation screen displays a "Confirm" button, which indicates that the contents of the image interpretation report have been confirmed.

The report creation device 36 creates an image interpretation report based on the input of the finding information and the like to the report creation screen. The content to be input to the image interpretation report may be selected by the user from multiple items displayed on the report creation screen. The image interpretation report describes the medical image used for diagnosis in the examination (hereinafter referred to as the key image) and the findings observed in the key image. The key image is accompanied by an examination ID and an image ID based on the order information. The key image may be associated with the image interpretation report using a method such as HarperLink. In addition, a mark or annotation may be superimposed on the key image to explain the findings.

When the radiologist wishes to confirm the content displayed on the report creation screen, he or she clicks the "Confirm" button. At this time, the report creation device 36 stores the created radiology report, for example, in a medical image management server device in the medical image management system 39. In addition, when the "Confirm" button is clicked, the report creation device 36 sends the created radiology report to the medical information processing device 1. Furthermore, in response to a request from the examining physician, the report creation device 36 sends the created radiology report to the electronic medical record system 33.

The medical image diagnostic device 37 is a device for acquiring medical images, such as an X-ray diagnostic device, an X-ray CT (Computed Tomography) device, and a magnetic resonance imaging device. A plurality of medical image diagnostic devices 37 may be connected via the communication network 31. The medical image diagnostic device 37 performs imaging of the subject in response to an operation by a technician or the like to start imaging based on an examination order from the radiology department information system 35. By performing imaging, the medical image diagnostic device 37 collects data indicating the state of the subject's internal tissues. Next, the medical image diagnostic device 37 generates a medical image based on the data. The medical image diagnostic device 37 transmits the generated medical image to the medical image management system 39 together with order information.

The medical image management system (PACS: Picture Archiving and Communication Systems) 39 is an information system that manages medical images such as X-ray CT images and MR images. The medical image management system 39 may manage not only image data after reconstruction but also data before reconstruction (raw data). The medical image management system 39 includes a medical image management server device. The medical image management server device is a computer device that executes processes related to the management of medical images. The medical image management server device is composed of one to multiple computer devices depending on the scale of the system. The medical image management server device is connected to the communication network 31. When the medical image management server device receives medical images and order information from the medical image diagnostic device 37, it associates the medical images with the order information and stores them. The medical image management server device also stores the interpretation report created by the report creation device 36 and manages them appropriately.

The Laboratory Information System (LIS) 41 is an information system that manages information in the clinical laboratory in the hospital, and includes a Laboratory Information Server, a Laboratory Information Terminal, and testing equipment. In the Laboratory Information System 41, clinical laboratory technicians perform specimen testing and physiological function testing based on order information sent from the electronic medical record system 33. The specimen testing may be outsourced to an external testing facility such as a hygiene laboratory. The Laboratory Information Server is a computer device that executes processing related to the management of information in clinical testing. The Laboratory Information Server is composed of one or more computers depending on the scale of the system. The Laboratory Information Terminal is used by technicians in the clinical laboratory, and one or more are provided depending on the scale of the system. The Laboratory Information Server and the Laboratory Information Terminal are connected to the communication network 31.

The testing equipment is equipment related to specimen testing, in which blood, urine, and other specimens obtained from patients are analyzed by clinical laboratory technicians, and physiological function testing, in which the patient's electroencephalogram, electrocardiogram, and the like are measured. The specimen testing includes, for example, pathological testing, blood and biochemical testing, general testing of urine and stool, immunoserological testing, microbiological testing, and tests related to blood transfusion and organ transplantation. The physiological function testing includes, for example, electroencephalogram testing, respiratory function testing, cardiac testing, and fundus photography testing. A plurality of testing equipment is installed according to the type of specimen testing and the type of physiological function testing. At least a portion of the specimen testing may be performed by an external testing institution. In this case, the clinical testing department information server device receives and stores the test results related to the specimen testing from the external testing institution. The clinical testing department information server device also transmits the test results to the electronic medical record system 33.

The other department system 43 is, for example, a surgery department system, a rehabilitation department system, or a dialysis department system. As with the radiology department information system 35 and clinical laboratory department information system 41 described above, the other department system 43 also creates surgery implementation reports, rehabilitation implementation reports, and the like in response to orders issued by doctors on electronic medical record terminal devices. The other department system 43 transmits the created surgery implementation reports, rehabilitation implementation reports, and the like to the electronic medical record system 33.

The medical information processing device 1 is connected to the hospital information system 3 via a communication interface 11 and a communication network 31. The medical information processing device 1 may be incorporated into the hospital information system 3. The medical information processing device 1 may also be used as a report creation device 36. The medical information processing device 1 shown in FIG. 1 has a communication interface 11, a memory 13 (storage unit), a display 15 (display unit), an input interface 17 (input unit), and a processing circuit 19 (processing unit). Data communication between the communication interface 11, the memory 13, the display 15, the input interface 17, and the processing circuit 19 is performed via a bus (BUS).

The communication interface 11 performs data communication with the hospital information system 3. The standard for communication between the communication interface 11 and the hospital information system 3 may be any standard, such as HL7 (Heart Level 7), DICOM, or both.

The memory 13 is a storage device such as a hard disk drive (HDD), a solid state drive (SSD), or an integrated circuit storage device that stores various information. The memory 13 stores data generated by executing, for example, the acquisition function 193, the identification function 195, the extraction function 197, the calculation function 199, the integration function 201, and the selection function 203. The generated data and these functions will be described later. The memory 13 stores multiple pieces of learning data generated by the integration function 201. The learning data is data that can be used for learning a machine learning model. The memory 13 may be a drive device that reads and writes various information between the memory 13 and a portable storage medium such as a compact disc (CD), a digital versatile disc (DVD), or a flash memory, or a semiconductor memory element such as a random access memory (RAM). The memory 13 also stores the control program according to this embodiment.

The display 15 displays various information. For example, the display 15 outputs a GUI (Graphical User Interface) for accepting various operations from the user. For example, the display 15 may be a liquid crystal display (LCD), a cathode ray tube (CRT), an organic electroluminescence display (OELD), a plasma display, or any other display, as appropriate. The display 15 may be a desktop type, or may be configured as a tablet terminal capable of wireless communication with the medical information processing device 1 main body.

The input interface 17 accepts various input operations from the user, converts the accepted input operations into electrical signals, and outputs them to the processing circuit 19. For example, the input interface 17 accepts selection instructions and various conditions from the user. As the input interface 17, for example, a mouse, keyboard, trackball, switch, button, joystick, touchpad, touch panel display, etc. can be used as appropriate.

In this embodiment, the input interface 17 is not limited to having physical operating parts such as a mouse, keyboard, trackball, switch, button, joystick, touchpad, and touch panel display. For example, an example of the input interface 17 includes an electrical signal processing circuit that receives an electrical signal corresponding to an input operation from an external input device provided separately from the device and outputs this electrical signal to the processing circuit 19. In addition, the input interface 17 may be configured as a tablet terminal or the like that is capable of wireless communication with the medical information processing device 1 main body.

The processing circuitry 19 controls the operation of the entire medical information processing device 1 in response to electrical signals of input operations output from the input interface 17. For example, the processing circuitry 19 has, as hardware resources, processors such as a CPU (Central Processing Unit), MPU (Micro Processing Unit), and GPU (Graphics Processing Unit), and memories such as a ROM (Read Only Memory) and RAM.

The processing circuit 19 may also be realized by an application specific integrated circuit (ASIC), a field programmable gate array (FPGA), other complex programmable logic devices (CPLDs), simple programmable logic devices (SPLDs), etc.

The processing circuit 19 has a system control function 191, an acquisition function 193, a specification function 195, an extraction function 197, a calculation function 199, an integration function 201, and a selection function 203. Each of the system control function 191, the acquisition function 193, the specification function 195, the extraction function 197, the calculation function 199, the integration function 201, and the selection function 203 is stored in the memory 13 in the form of a program executable by a computer. The processing circuit 19 executes the system control function 191, the acquisition function 193, the specification function 195, the extraction function 197, the calculation function 199, the integration function 201, and the selection function 203 by a processor that executes the programs deployed in the memory.

That is, the processing circuit 19 corresponds to a processor that realizes a function corresponding to each program by reading and executing the program from the memory 13. In other words, the processing circuit 19 in a state in which each program has been read has a function corresponding to the read program. Note that each function (191, 193, 195, 197, 199, 201, 203) is not limited to being realized by a single processing circuit. A processing circuit may be configured by combining multiple independent processors, and each processor may realize each function (191, 193, 195, 197, 199, 201, 203) by executing a program. The processing circuit 19 that realizes the system control function 191, acquisition function 193, identification function 195, extraction function 197, calculation function 199, integration function 201, and selection function 203 is an example of a system control unit, acquisition unit, identification unit, extraction unit, calculation unit, integration unit, and selection unit.

The system control function 191 controls each function of the processing circuit 19 based on an input operation received from a user via the input interface 17. Specifically, the system control function 191 reads out a control program stored in the memory 13, expands it on the memory in the processing circuit 19, and controls each part of the medical information processing device 1 according to the expanded control program.

The acquisition function 193 acquires the first examination results from the radiology department information system 35 when the first examination results regarding the medical images acquired by the first examination on the subject are completed. The first examination corresponds to an image examination by the medical image diagnostic device 37. The first examination results correspond to an image interpretation report. For example, when the contents of the image interpretation report are confirmed by the image interpretation physician by the acquisition function 193, the processing circuitry 19 acquires the image interpretation report via the communication interface 11. Based on the acquired image interpretation report, the processing circuitry 19 acquires order information regarding the image interpretation report from the electronic medical record system 33.

The acquisition function 193 acquires the second test results of the second test performed on the subject for the test purpose of the first test from the medical data server based on the order information related to the first test results. The test purpose of the first test is identified based on the acquired first test results by the identification function 195 described below. The second test is, for example, a specimen test or a physiological function test. In this case, the medical data server corresponds to a clinical testing department information server device or an electronic medical record server device. In the following, for the sake of concrete explanation, the specimen test is assumed to be a pathological test or a blood/biochemical test. In this case, the second test results correspond to the clinical test reports of each of the pathological test, blood/biochemical test, and physiological function test. The processing circuit 19 acquires the electronic medical record of the subject from the electronic medical record server device, which is a type of medical data server, based on the order information by the acquisition function 193.

The identification function 195 identifies the examination purpose of the first examination based on the first examination results related to the medical images acquired by the first examination on the subject. Specifically, the processing circuitry 19 performs natural language processing based on a predefined dictionary on the order information acquired by the identification function 195. As a result, the processing circuitry 19 identifies the examination purpose from the order information. The processing circuitry 19 assigns the time when the examination order was issued, the order number, and the patient ID to the identified examination purpose, and also assigns a data ID that is a unique ID specific to this medical information processing device 1.

The extraction function 197 extracts first examination data that can be used for learning the machine learning model from the first examination result acquired by the acquisition function 193. The first examination data has information indicating the interpretation result, such as, for example, the lesion characteristics in the interpretation report, the size of the lesion, the suspicion of disease, the corresponding next action, and past examination results. Specifically, the processing circuitry 19 performs natural language processing based on a predefined dictionary on the information associated with the hyperlink or medical image in the interpretation report by the extraction function 197. In this way, the processing circuitry 19 extracts the first examination data from the first examination result. Note that the extraction target by the extraction function 197 may be predetermined for each hierarchical level in the structured interpretation report. The extraction target may also be selected by the user via the input interface 17. The processing circuitry 19 may also display the contents of the extracted first examination data on the display 15. At this time, the user selects a desired item to be used as learning data from the contents of the first examination data via the input interface 17.

The extraction function 197 extracts learning images that can be used for the learning from the medical images in the first examination results acquired by the acquisition function 193. Specifically, the processing circuitry 19 extracts a key image related to the image interpretation report from the image interpretation report using the extraction function 107. The processing circuitry 19 may cut out the key image at a preset size. The processing circuitry 19 may also extract an area in the key image that is specified by the image interpretation physician or that is set by a predetermined image processing at a preset size. The processing circuitry 19 may also extract a key image with a mark or annotation written thereon at a preset size. The extracted key image, i.e., the medical image, corresponds to the learning image.

The processing circuitry 19 uses the extraction function 197 to assign to the learning image a system examination ID and a system image ID that are different from the examination ID and image ID assigned to the key image. The learning image corresponds to a lesion image. The processing circuitry 19 temporarily stores the lesion image with the system examination ID and system image ID assigned to it in the memory 13.

The extraction function 197 extracts tag information such as the type of modality, manufacturer, device name, device ID, protocol information (protocol name, procedure step information, etc.), modality-specific imaging parameters (exposure information, compression rate, tube voltage, tube current, etc., items described in DICOM Part 3 C.8), image reconstruction parameters, contrast information, examination area, number of detector rows, type, model number, etc., from the DICOM tag (including private tags) attached to the medical image that is the source of the key image. Note that if additional imaging of the subject or imaging by another medical image diagnostic device 37 is ordered in the order information, the processing circuitry 19 repeats the extraction of the first examination data, learning images, and tag information. At this time, the processing circuitry 19 extracts (calculates) the examination interval (period) in the multiple first examinations.

The extraction function 197 extracts second test data that can be used for the learning from the second test results acquired by the acquisition function 193. Specifically, the processing circuit 19 uses the extraction function 197 to extract second test results of the second test related to the first test from the clinical laboratory department information system 41 or other department systems 43 based on the patient information and the order number and test purpose in the order information. More specifically, the processing circuit 19 extracts second test results that have the same patient name, order number, test purpose, and electronic medical record record in the first test from the clinical laboratory department information system 41 or other department systems 43. Note that, if an additional specimen test for the subject is ordered in the order information, the processing circuit 19 calculates the test interval (period) in multiple second tests. In the following, for the sake of concrete explanation, when the second test is a specimen test, the second test result corresponds to a specimen test report. In this case, the second test data corresponds to a specimen test result.

If the specimen test is a pathological test, the second test corresponds to the pathological test, and the second test result corresponds to the tissue diagnosis report. In the pathological test of cytology, the type of cancer, the malignancy of the cancer, the proliferation ability of the cancer cells, the presence or absence of hormone receptors, the overexpression of HER2 protein, the presence or absence of HER2 gene amplification, etc. contribute to the diagnosis. In the pathological test of biopsy, the distribution state and shape, the type classification, the presence or absence of malignant cells at the resection border, the size, the presence or absence of infiltration, etc. contribute to the diagnosis. The results of the pathological test are described in the tissue diagnosis report for the above items that contribute to the diagnosis. The processing circuit 19 performs natural language processing based on a predefined dictionary using the extraction function 197 on the tissue diagnosis report. As a result, the processing circuit 19 extracts the pathological test results for the items that contribute to the diagnosis from the tissue diagnosis report as the second test data.

In addition, if the specimen test is a blood and biochemistry test, the second test corresponds to the blood and biochemistry test, and the second test result corresponds to the blood and biochemistry test report. The processing circuit 19 uses the extraction function 197 to extract the numerical values for the various items in the blood and biochemistry test, i.e., the blood and biochemistry test results, from the blood and biochemistry test report as the second test data.

The extraction function 197 extracts the subject's diagnosis result based on the first test result and the second test result from the electronic medical record acquired by the acquisition function 193. Specifically, the processing circuitry 19 uses the extraction function 197 to perform natural language processing on the electronic medical record based on a predefined dictionary. As a result, the processing circuitry 19 extracts the diagnosis result from the electronic medical record. Note that the source from which the diagnosis result is extracted may be a part of the electronic medical record related to the order information.

The extraction function 197 extracts treatment results corresponding to the diagnosis results from the electronic medical record acquired by the acquisition function 193. Specifically, the processing circuitry 19 executes natural language processing on the electronic medical record based on a predefined dictionary using the extraction function 197. In this way, the processing circuitry 19 extracts the treatment results from the electronic medical record. Note that the source from which the treatment results are extracted may be a part of the electronic medical record related to the diagnosis result.

The extraction function 197 extracts patient information on the subject from the electronic medical record acquired by the acquisition function 193. The patient information includes, for example, height, weight, BMI, blood pressure, body fat, sex, age, medical history, ethnicity (race), genomic information (for example, genetic diagnosis results), occupation, dietary habits, drinking history, smoking history, exercise habits, family history, and in the case of women, childbirth history, age at menarche, age at menopause, menstrual status, etc. Specifically, the processing circuit 19 performs natural language processing on the electronic medical record based on a predefined dictionary using the extraction function 197. In this way, the processing circuit 19 extracts patient information from the electronic medical record. Note that the source of patient information is not limited to the electronic medical record, and may be a medical questionnaire, health check information, etc. on the subject. The medical questionnaire and health check information are linked to the electronic medical record, for example.

Figure 2 is a diagram showing an example of the relationship between the acquisition target by the acquisition function 193 and the extraction target by the extraction function 197. As shown in Figure 2, when an image interpretation report is acquired, the purpose of the examination is identified from the examination order in the image interpretation report. Next, extraction targets such as examination results, diagnosis results, and treatment results related to the same examination purpose as the image interpretation report are extracted.

The calculation function 199 calculates the test reliability indicating the reliability (or credibility) of the first test result based on the first test result and the diagnosis result. The test reliability is, for example, a numerical representation of the specificity (true positive, true negative, false positive, false negative) of the first test result as a value from 0 to 1 or a percentage. The processing circuit 19 calculates, for example, the agreement rate between the diagnosis result and the findings in the image interpretation report as the test reliability using the calculation function 199. The processing circuit 19 calculates, for example, the diagnostic reliability indicating the reliability (or credibility) of the diagnosis result based on the diagnosis result and the treatment result. The diagnostic reliability is, for example, a numerical representation of the specificity (true positive, true negative, false positive, false negative) of the diagnosis result as a value from 0 to 1 or a percentage. The processing circuit 19 calculates, for example, the agreement rate between the treatment result and the diagnosis result as the diagnostic reliability.

The integration function 201 integrates the data extracted by the extraction function 197 and the examination purpose identified by the identification function 195 based on the order number and the patient ID as learning data that can be used for learning. Specifically, the processing circuit 19 generates learning data by integrating the first examination data, the second examination data, the learning image, and the examination purpose using the integration function 201. More specifically, the second examination data is weighted in advance according to the type of the second examination. The weight may be set individually according to clinical types such as the examination site and the examination purpose. Since the results of the pathological examination contribute most to the diagnosis, the weight related to the pathological examination is the largest among the second examinations by default. The weight is displayed on the display 15 and can be changed by a user's instruction via the input interface 17.

The processing circuit 19 assigns the same data ID as the test purpose to the learning data by the integration function 201. When selecting learning data, the data ID is used instead of the patient name from the viewpoint of anonymity and security. The processing circuit 19 may further integrate at least one of the patient information, test reliability, diagnosis result, treatment result, and diagnosis reliability into the learning data.

The processing circuitry 19 may also use the integration function 201 to integrate procedure names and the like in the surgery report into the learning data. At this time, the processing circuitry 19 first uses the acquisition function 193 to acquire the surgery report from the surgery department system based on the order information. Next, the processing circuitry 19 uses the extraction function 197 to extract procedure names and the like from the surgery report using natural language processing. Next, the processing circuitry 19 uses the integration function 201 to integrate the procedure names and the like into the learning data.

The learning data is stored in the memory 13 in a text format such as CSV (Comma Separated Values). The storage format of the learning data is not limited to CSV, and may be other storage formats such as a database format or an XML (extensible Markup Language) format. The generation of the learning data may be updated again as appropriate after the electronic medical record, which is the final stage of medical treatment for the subject, is finalized. For example, after the electronic medical record is finalized, all test results may be extracted at once after the medical record is finalized, in response to a user instruction via the input interface 17. At this time, the user appropriately updates the learning data by inputting via the input interface 17.

Figure 3 is a diagram showing an example of learning data generated by integrating various data by the integration function 201. As shown in Figure 3, the processing circuitry 19 automatically generates learning data to which a data ID is assigned by integrating the data extracted by the extraction function 197, the examination purpose identified by the identification function 195, and the calculation result calculated by the calculation function 199 by the integration function 201. Of the multiple weights shown in Figure 3, the weight assigned to the pathological examination result is the largest. The processing circuitry 19 stores the generated learning data in the memory 13. The processing circuitry 19 stores the learning data in the memory 13 as described above in response to the completion of the image interpretation report. That is, the memory 13 stores a group of learning data (hereinafter referred to as a learning data set).

The selection function 203 selects learning data to be used for learning from a group of learning data (learning data set) stored in the memory 13 based on at least one of the test reliability, the diagnostic reliability, and the learning content for the machine learning model. For example, in accordance with a user's instruction via the input interface 17, the processing circuit 19 selects cases and learning data desired by the user using the selection function 203. Specifically, the processing circuit 19 applies a selection method such as a filter set by the user to the learning data set to select target and non-target data for learning. The processing circuit 19 may format the selected learning data in accordance with a learning system (not shown) and output it in a specific form. The output format of the learning data can be selected by the user in accordance with the input method of the learning system. The input method of the learning system is, for example, CSV or text, and image data, a database method, or an XML method.

The above describes the overall configuration of the medical information processing device 1 and the hospital information system 3. Below, the processing procedure of the learning data generation process that automatically generates learning data will be described. To make the explanation more specific, it is assumed that the first test result is an image interpretation report and the second test result is a specimen test report. Figure 4 is a diagram showing an example of the processing procedure of the learning data generation process.

(Learning data generation process)
(Step S401)
The processing circuitry 19, in response to clicking the "Confirm" button on the report creation screen, i.e., upon completion of the image interpretation report corresponding to the first examination result, acquires the image interpretation report from the medical image management system 39 using the acquisition function 193. Based on the acquired image interpretation report, the processing circuitry 19 acquires order information related to the image interpretation report from the electronic medical record system 33. The acquired image interpretation report and order information are temporarily stored in the memory 13.

(Step S402)
The processing circuitry 19 identifies the examination purpose in the order information by the identification function 195. Specifically, the processing circuitry 19 identifies the examination purpose described in the order information by applying natural language processing to the order information. The processing circuitry 19 assigns a data ID to the identified examination purpose. The examination purpose to which the data ID is assigned is temporarily stored in the memory 13.

(Step S403)
The processing circuitry 19 uses the acquisition function 193 to acquire a specimen testing report according to the specified testing purpose from the clinical laboratory information system 41. For example, the processing circuitry 19 acquires a specimen testing report having the same patient ID as the patient ID described in the order information and the same testing purpose as the specified testing purpose from a database in the clinical laboratory information system 41. The processing circuitry 19 uses the acquisition function 193 to acquire the electronic medical record of the subject from the electronic medical record system 33 based on the order information. The acquired assay test report and electronic medical record are temporarily stored in the memory 13.

(Step S404)
The processing circuitry 19 extracts patient information, diagnosis results, and treatment results from the electronic medical record by the extraction function 197. If the diagnosis results and treatment results are not reflected in the electronic medical record, the processing circuitry 19 extracts the diagnosis results and treatment results from the electronic medical record acquired after a predetermined period of time has elapsed, for example, the day before the re-examination date and time of the subject. The processing circuitry 19 extracts the interpretation results and the lesion image from the image interpretation report, and extracts the specimen examination results from the specimen examination report. The extracted patient information, diagnosis results, treatment results, image interpretation results, lesion image, and specimen examination results are temporarily stored in the memory 13. If the first and second examinations are further scheduled, the processing circuitry 19 repeats the processes of steps S401 to S404 after the dates and times of these examinations.

(Step S405)
The processing circuitry 19 calculates the examination reliability based on the image interpretation report and the diagnosis result by the calculation function 199. The processing circuitry 19 also calculates the diagnostic reliability based on the diagnosis result and the treatment result. The calculated examination reliability and diagnostic reliability are temporarily stored in the memory 13.

(Step S406)
The processing circuitry 19 integrates the patient information, diagnosis results, treatment results, image interpretation results, lesion images, specimen test results, test reliability, and diagnostic reliability as learning data by the integration function 201. Next, the processing circuitry 19 assigns a system ID to the learning data. The processing circuitry 19 stores the learning data to which the system ID has been assigned in the memory 13. The processing circuitry 19 repeats the processes of steps S401 to S406 according to the completion of the image interpretation report. As a result, the memory 13 stores a group of learning data (learning data set) according to the total number of image interpretation reports. At this time, the memory 13 may function as a database of learning data. This step ends the learning data generation process.

When using a group of learning data stored in memory 13, processing circuit 19 uses selection function 203 to select learning data to be used for learning from the group of learning data stored in memory 13 based on at least one of the test reliability, diagnostic reliability, and learning content for the machine learning model. For example, when learning is performed on a machine learning model that inputs lesion images and outputs diagnostic results, processing circuit 19 uses selection function 203 to select learning data with high diagnostic reliability and test reliability.

According to the medical information processing device 1 according to the embodiment described above, the examination purpose of the first examination is identified based on the first examination result related to the medical image acquired by the first examination on the subject, the second examination result of the second examination performed on the subject for the examination purpose is acquired from the medical data server based on the order information related to the first examination result, first examination data usable for learning the machine learning model is extracted from the first examination result, second examination data usable for the learning is extracted from the second examination result, learning images usable for the learning are extracted from the medical image, and the first examination data, second examination data, learning image, and examination purpose are integrated as learning data usable for the learning. As a result, according to the medical information processing device 1, learning data can be automatically generated when the first examination result is acquired, without bothering the image interpretation doctor.

In addition, according to the medical information processing device 1, the electronic medical record of the subject is further obtained from the medical data server based on the order information, and at least one of the diagnosis result of the subject based on the first test result and the second test result, the treatment result corresponding to the diagnosis result, and the patient information is extracted from the electronic medical record, and at least one of the diagnosis result, the treatment result, and the patient information is further integrated into the learning data. In addition, according to the medical information processing device 1, the test reliability indicating the reliability of the first test result is calculated based on the first test result and the diagnosis result, and the test reliability is further integrated into the learning data. In addition, according to the medical information processing device 1, the diagnostic reliability indicating the degree of reliability of the diagnostic result is calculated based on the diagnosis result and the treatment result, and the diagnostic reliability is further integrated into the learning data. As a result, according to the medical information processing device 1, at least one of the diagnosis result, the treatment result, the patient information, the test reliability, and the diagnostic reliability can be automatically further integrated into the learning data without bothering the user.

As described above, according to the present medical information processing device 1, various data (second examination results) related to the training data can be automatically collected based on the radiology doctor's interpretation report (first examination result) during normal image interpretation work. This eliminates the need for the radiologist to check the second examination results, which are medical documents such as pathology examination results and medical records, when creating the training data, reducing the burden and labor of the radiologist and improving the efficiency of creating the training data. In addition, according to the present medical information processing device 1, diagnosis results, treatment results, patient information, examination reliability, and diagnosis reliability can be further used in selecting the training data, and more accurate learning can be performed on the machine learning model.

(First application example)
The first application example is to update the past test reliability (hereinafter referred to as past test reliability) for the same test purpose as the test purpose to the calculated test reliability based on the treatment result. The first application example corresponds to a situation where a repeat examination or follow-up (observation, follow-up survey) is performed on the same subject after one set of examination, diagnosis, and treatment is performed on the subject.

The processing circuit 19 uses the acquisition function 193 to acquire past learning data (hereinafter referred to as past learning data) related to the same test purpose as the identified test purpose and the same patient information as the extracted patient information from the learning dataset. The processing circuit 19 uses the acquisition function 193 to acquire the past test reliability from the past learning data. The processing circuit 19 uses the extraction function 197 to perform natural language processing on the acquired electronic medical record based on a predefined dictionary, and extracts the diagnosis, stage, severity of the disease, and other status changes, and whether or not treatment has been performed.

Based on the extracted information, the processing circuit 19 performs recalculation of the past diagnostic reliability, which indicates the credibility of the past test, using the calculation function 199. When recalculating the past test reliability, the display 15 may display whether or not the test reliability needs to be reviewed, i.e., whether or not recalculation is required. At this time, the user selects whether or not review is required via the input interface 17. The processing circuit 19 recalculates the past test reliability by adding or subtracting a predetermined value to the past test reliability based on the treatment results and the diagnosis results in the past learning data (hereinafter referred to as past diagnosis results). The predetermined value may be stored in the memory 13 in advance, or may be set by a user such as a doctor or administrator via the input interface 17. The predetermined value may be different for addition and subtraction.

Specifically, if a comparison between the treatment results and the past diagnosis results shows that the treatment has reduced the stage and severity of the disease, i.e., if the past diagnosis results are determined to be valid, the processing circuit 19 adds a predetermined value to the past examination reliability using the calculation function 199. Also, if a comparison between the treatment results and the past diagnosis results shows that the treatment has not reduced the stage and severity of the disease or has worsened the stage and severity of the disease, the processing circuit 19 subtracts a predetermined value from the past examination reliability. Also, if a comparison between the treatment results and the past diagnosis results shows that the diagnosis name has changed, the processing circuit 19 subtracts a predetermined value from the past examination reliability.

The display 15 may also display a comparison between the treatment results and past diagnosis results and the reliability of past tests. At this time, the reliability of past tests may be changed according to a user instruction via the input interface 17.

The processing circuit 19 updates the past inspection reliability to the recalculated inspection reliability by the integration function 201. The processing circuit 19 updates the past learning data by integrating the updated past inspection reliability into the past learning data. The processing circuit 19 incorporates the updated past learning data into the learning data set and stores it in the memory 13.

According to the medical information processing device 1 of the first application example of the embodiment described above, past learning data having past test reliability for the same test purpose as the test purpose is acquired, and a predetermined value is added or subtracted from the past test reliability based on the treatment result and the diagnosis result in the past learning data, thereby recalculating the past test reliability and updating the past test reliability to the recalculated test reliability. For example, when the correlation between the treatment result and the past diagnosis result is low and the past test reliability is high, the credibility of the interpretation result in the interpretation report is low. At this time, according to the present medical information processing device 1, the test reliability indicating the credibility of the interpretation result can be reduced. Also, when the correlation between the treatment result and the past diagnosis result is high and the past test reliability is high, the credibility of the interpretation result in the interpretation report is high. At this time, according to the present medical information processing device 1, the test reliability indicating the credibility of the interpretation result can be increased.

From the above, the medical information processing device 1 according to the first application example can adjust the reliability of past tests according to the effectiveness of the treatment results and the accuracy of the diagnosis in the past diagnosis results. In addition, the medical information processing device 1 can improve the accuracy of the learning data, and can perform more accurate learning for the machine learning model.

(Second Application Example)
The second application example is to update the past diagnostic reliability (hereinafter referred to as past diagnostic reliability) for the same test purpose as the test purpose to the calculated diagnostic reliability based on the treatment result. The second application example corresponds to a situation where a repeat examination or follow-up (observation, follow-up survey) is performed on the same test subject after one set of tests, diagnosis, and treatment is performed on the subject.

The processing circuitry 19 acquires past learning data from the learning dataset using the acquisition function 193. The processing circuitry 19 acquires the electronic medical record of the same subject as the subject related to the treatment result from the electronic medical record server device. The processing circuitry 19 extracts, using the extraction function 197, from the acquired electronic medical record, whether there has been any change in condition such as the diagnosis, stage, and severity of the disease, and whether or not treatment has been performed.

Based on the extracted information, the processing circuit 19 performs recalculation of the past diagnostic reliability, which indicates the credibility of the past diagnosis, using the calculation function 199. When recalculating the past diagnostic reliability, the display 15 may display whether or not the diagnostic reliability needs to be reviewed, i.e., whether or not recalculation is required. At this time, the user selects whether or not review is required via the input interface 17. The processing circuit 19 recalculates the past diagnostic reliability by adding or subtracting a predetermined value to the past diagnostic reliability based on the treatment results and the diagnostic results in the past learning data (hereinafter referred to as past diagnostic results). The predetermined value may be stored in the memory 13 in advance, or may be set by a user such as a doctor or administrator via the input interface 17. The predetermined value may be different for addition and subtraction.

Specifically, if a comparison between the treatment results and the past diagnosis results shows that the treatment has reduced the stage and severity of the disease, i.e., if the past diagnosis results are determined to be valid, the processing circuit 19 adds a predetermined value to the past diagnosis reliability using the calculation function 199. Also, if a comparison between the treatment results and the past diagnosis results shows that the treatment has not reduced the stage and severity of the disease or has worsened the stage and severity of the disease, the processing circuit 19 subtracts a predetermined value from the past diagnosis reliability. Also, if a comparison between the treatment results and the past diagnosis results shows that the diagnosis name has changed, the processing circuit 19 subtracts a predetermined value from the past diagnosis reliability.

The display 15 may also display a comparison between the treatment results and past diagnosis results and the reliability of the past diagnosis. At this time, the reliability of the past diagnosis may be changed by a user instruction via the input interface 17.

The processing circuit 19 uses the integration function 201 to update the past diagnostic reliability to the recalculated diagnostic reliability. The processing circuit 19 updates the past learning data by integrating the updated past diagnostic reliability into the past learning data. The processing circuit 19 incorporates the updated past learning data into the learning data set and stores it in the memory 13.

According to the medical information processing device 1 of the second application example of the embodiment described above, past learning data having a past diagnostic reliability for the same test purpose as the test purpose is acquired, and a predetermined value is added or subtracted from the past diagnostic reliability based on the treatment result and the diagnostic result in the past learning data, thereby recalculating the past diagnostic reliability and updating the past diagnostic reliability to the recalculated diagnostic reliability. For example, when the correlation between the treatment result and the past diagnostic result is low, the reliability of the diagnostic result is low. At this time, according to the present medical information processing device 1, the diagnostic reliability indicating the reliability of the diagnostic result can be reduced. Also, when the correlation between the treatment result and the past diagnostic result is high, the reliability of the diagnostic result is high. At this time, according to the present medical information processing device 1, the test reliability indicating the reliability of the diagnostic result can be increased.

From the above, the medical information processing device 1 according to the second application example can adjust the reliability of past diagnoses according to the effectiveness of the treatment results and the accuracy of the diagnosis in the past diagnosis results. In addition, the medical information processing device 1 can improve the accuracy of the learning data, and can perform more accurate learning for the machine learning model.

When the technical idea of this embodiment is realized by a learning data generation program, the computer is made to identify the examination purpose of the first examination based on the first examination results related to the medical images acquired by the first examination on the subject, acquire the second examination results of the second examination performed on the subject for the examination purpose from the medical data server based on the order information related to the first examination results, extract first examination data usable for learning the machine learning model from the first examination results, extract second examination data usable for the learning from the second examination results, extract learning images usable for the learning from the medical images, and integrate the first examination data, second examination data, and learning images as learning data usable for the learning.

For example, the learning data generation process can be realized by installing a learning data generation program in a computer in the hospital information system 3 and expanding it in memory. In this case, the program that can cause the computer to execute the method can also be stored and distributed on a storage medium such as a magnetic disk (such as a hard disk), an optical disk (such as a CD-ROM or DVD), or a semiconductor memory. The processing procedure and effects of the learning data generation program are the same as those in the embodiment, so a description thereof will be omitted.

When the technical idea of this embodiment is realized by a training data generation method, the training data generation method identifies the examination purpose of the first examination based on the first examination result related to the medical image acquired by the first examination on the subject, acquires the second examination result of the second examination performed on the subject for the examination purpose from the medical data server based on the order information related to the first examination result, extracts first examination data usable for learning the machine learning model from the first examination result, extracts second examination data usable for the learning from the second examination result, extracts training images usable for the learning from the medical image, and integrates the first examination data, the second examination data, and the training images as training data usable for the learning. The processing procedure and effects of the training data generation method are similar to those of the embodiment, and therefore will not be described.

The above-described embodiments and application examples make it possible to easily generate training data.

Although several embodiments of the present invention have been described, these embodiments are presented as examples and are not intended to limit the scope of the invention. These embodiments can be implemented in various other forms, and various omissions, substitutions, and modifications can be made without departing from the gist of the invention. These embodiments and their modifications are within the scope of the invention and its equivalents as set forth in the claims, as well as the scope and gist of the invention.

REFERENCE SIGNS LIST 1 Medical information processing device 3 Hospital information system 11 Communication interface 13 Memory 15 Display 17 Input interface 19 Processing circuit 31 Communication network 33 Electronic medical record system 35 Radiology department information system 36 Report creation device 37 Medical image diagnostic device 39 Medical image management system 41 Clinical testing department information system 43 Other department systems 191 System control function 193 Acquisition function 195 Identification function 197 Extraction function 199 Calculation function 201 Integration function 203 Selection function

Claims (11)

an identification unit that identifies an examination purpose of the first examination based on a first examination result regarding a medical image acquired by the first examination on a subject;
an acquisition unit that acquires, from a medical data server, second test results of a second test performed on the subject for the test purpose based on order information related to the first test results;
an extraction unit that extracts first test data usable for learning a machine learning model from the first test result, extracts second test data usable for the learning from the second test result, extracts learning images usable for the learning from the medical images , and extracts a diagnosis result of the subject based on the first test result and the second test result from the electronic medical record of the subject ;
an integration unit that integrates the first test data, the second test data, the learning image, the test purpose, and the diagnosis result as learning data that can be used for the learning;
A medical information processing device comprising: The acquiring unit further acquires an electronic medical record of the subject from the medical data server based on the order information .
The medical information processing device according to claim 1 . a calculation unit that calculates a test reliability indicating the reliability of the first test result based on the first test result and the diagnosis result,
The integration unit further integrates the test reliability into the learning data.
The medical information processing device according to claim 1 . The extraction unit extracts, from the electronic medical record, a treatment result for the subject corresponding to the diagnosis result;
The integration unit further integrates the treatment result into the learning data.
The medical information processing device according to claim 3 . The acquisition unit acquires past learning data having past test reliability for the same test purpose as the test purpose,
the calculation unit recalculates the past test reliability by adding or subtracting a predetermined value to the past test reliability based on the treatment result and the diagnosis result in the past learning data;
The integration unit updates the past test reliability to the recalculated test reliability.
The medical information processing device according to claim 4 . the calculation unit calculates a diagnostic reliability indicating a degree of reliability of the diagnostic result based on the diagnostic result and the treatment result;
The integration unit further integrates the diagnostic reliability into the learning data.
The medical information processing device according to claim 4 . The acquisition unit acquires past learning data having a past diagnostic reliability for the same test purpose as the test purpose,
the calculation unit recalculates the past diagnostic reliability by adding or subtracting a predetermined value to the past diagnostic reliability based on the treatment result and the diagnosis result in the past learning data;
The integration unit updates the past diagnostic reliability to the recalculated diagnostic reliability.
The medical image processing device according to claim 4 . A memory that stores a plurality of the learning data;
and a selection unit that selects learning data to be used for the learning from a group of the learning data stored in the memory based on at least one of the test reliability, the diagnostic reliability, and a learning content for the machine learning model.
The medical information processing device according to claim 6 . The extraction unit extracts patient information related to the subject from at least one of the electronic medical record, the medical questionnaire, and the medical examination information,
The integration unit further integrates the patient information in the learning data.
The medical image processing device according to claim 2 . On the computer,
Identifying an examination purpose of the first examination based on a first examination result regarding a medical image acquired by the first examination on the subject;
obtaining, from a medical data server, second test results of a second test performed on the subject for the test purpose based on order information related to the first test results;
Extracting first inspection data usable for learning a machine learning model from the first inspection result, extracting second inspection data usable for the learning from the second inspection result, and extracting learning images usable for the learning from the medical images;
extracting a diagnosis of the subject based on the first test result and the second test result from an electronic medical record of the subject;
Integrating the first test data, the second test data, the learning image, and the diagnosis result as learning data available for the learning;
A learning data generation program that makes this possible. The computer
Identifying an examination purpose of the first examination based on a first examination result regarding a medical image acquired by the first examination on the subject;
obtaining, from a medical data server, second test results of a second test performed on the subject for the test purpose based on order information related to the first test results;
Extracting first inspection data usable for learning a machine learning model from the first inspection result, extracting second inspection data usable for the learning from the second inspection result, and extracting learning images usable for the learning from the medical images;
extracting a diagnosis of the subject based on the first test result and the second test result from an electronic medical record of the subject;
A learning data generating method for integrating the first test data, the second test data, the learning image, and the diagnosis result as learning data usable for the learning.

Images